Restoring zero initial scale behavior

charts/kserve-resources/templates/clusterrole.yaml

@@ -186,14 +186,6 @@ rules:
   - get
   - patch
   - update
-- apiGroups:
-  - operator.knative.dev
-  resources:
-  - knativeservings
-  verbs:
-  - get
-  - list
-  - watch
 - apiGroups:
   - rbac.authorization.k8s.io
   resourceNames:

cmd/manager/main.go

@@ -31,7 +31,6 @@ import (
 	typedcorev1 "k8s.io/client-go/kubernetes/typed/core/v1"
 	_ "k8s.io/client-go/plugin/pkg/client/auth/gcp"
 	"k8s.io/client-go/tools/record"
-	operatorv1beta1 "knative.dev/operator/pkg/apis/operator/v1beta1"
 	knservingv1 "knative.dev/serving/pkg/apis/serving/v1"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client/config"
@@ -191,19 +190,6 @@ func main() {
 		}
 	}
 
-	knServingFound, knServingCheckErr := utils.IsCrdAvailable(cfg, operatorv1beta1.SchemeGroupVersion.String(), constants.KnativeServingKind)
-	if knServingCheckErr != nil {
-		setupLog.Error(knServingCheckErr, "error when checking if Knative KnativeServing kind is available")
-		os.Exit(1)
-	}
-	if knServingFound {
-		setupLog.Info("Setting up Knative Operator scheme")
-		if err := operatorv1beta1.AddToScheme(mgr.GetScheme()); err != nil {
-			setupLog.Error(err, "unable to add Knative Operator APIs to scheme")
-			os.Exit(1)
-		}
-	}
-
 	if !ingressConfig.DisableIstioVirtualHost {
 		vsFound, vsCheckErr := utils.IsCrdAvailable(cfg, istioclientv1beta1.SchemeGroupVersion.String(), constants.IstioVirtualServiceKind)
 		if vsCheckErr != nil {

config/configmap/inferenceservice.yaml

@@ -59,7 +59,6 @@ data:
         # revisions, which prevents the reconciliation loop to be triggered if the annotations is 
         # configured here are used.
         # Default values are:
-        #  "autoscaling.knative.dev/initial-scale",
         #  "autoscaling.knative.dev/min-scale",
         #  "autoscaling.knative.dev/max-scale",
         #  "internal.serving.kserve.io/storage-initializer-sourceuri",

config/overlays/odh/inferenceservice-config-patch.yaml

@@ -88,7 +88,6 @@ data:
   inferenceService: |-
     {
       "serviceAnnotationDisallowedList": [
-        "autoscaling.knative.dev/initial-scale",
         "autoscaling.knative.dev/min-scale",
         "autoscaling.knative.dev/max-scale",
         "internal.serving.kserve.io/storage-initializer-sourceuri",

config/overlays/test/configmap/inferenceservice-openshift-ci-serverless-predictor.yaml

@@ -88,7 +88,6 @@ data:
   inferenceService: |-
     {
       "serviceAnnotationDisallowedList": [
-        "autoscaling.knative.dev/initial-scale",
         "autoscaling.knative.dev/min-scale",
         "autoscaling.knative.dev/max-scale",
         "internal.serving.kserve.io/storage-initializer-sourceuri",

config/rbac/role.yaml

@@ -173,14 +173,6 @@ rules:
   - get
   - patch
   - update
-- apiGroups:
-  - operator.knative.dev
-  resources:
-  - knativeservings
-  verbs:
-  - get
-  - list
-  - watch
 - apiGroups:
   - rbac.authorization.k8s.io
   resourceNames:

go.mod

@@ -44,7 +44,6 @@ require (
 	k8s.io/kube-openapi v0.0.0-20250304201544-e5f78fe3ede9
 	k8s.io/utils v0.0.0-20241210054802-24370beab758
 	knative.dev/networking v0.0.0-20250117155906-67d1c274ba6a
-	knative.dev/operator v0.42.2
 	knative.dev/pkg v0.0.0-20250117084104-c43477f0052b
 	knative.dev/serving v0.44.0
 	sigs.k8s.io/controller-runtime v0.19.1

go.sum

@@ -397,8 +397,6 @@ github.com/linode/linodego v1.40.0 h1:7ESY0PwK94hoggoCtIroT1Xk6b1flrFBNZ6KwqbTql
 github.com/linode/linodego v1.40.0/go.mod h1:NsUw4l8QrLdIofRg1NYFBbW5ZERnmbZykVBszPZLORM=
 github.com/mailru/easyjson v0.9.0 h1:PrnmzHw7262yW8sTBwxi1PdJA3Iw/EKBa8psRf7d9a4=
 github.com/mailru/easyjson v0.9.0/go.mod h1:1+xMtQp2MRNVL/V1bOzuP3aP8VNwRW55fQUto+XFtTU=
-github.com/manifestival/manifestival v0.7.2 h1:l4uFdWX/xQK4QcRfqGoMtBvaZeWPEuwD6hVsCwUqZY4=
-github.com/manifestival/manifestival v0.7.2/go.mod h1:nl3T6HlfHCeidooWVTMI9vYNTBkQ1GdhLNb+smozbdk=
 github.com/mattn/go-colorable v0.1.13 h1:fFA4WZxdEF4tXPZVKMLwD8oUnCTTo08duU7wxecdEvA=
 github.com/mattn/go-colorable v0.1.13/go.mod h1:7S9/ev0klgBDR4GtXTXX8a3vIGJpMovkB8vQcUbaXHg=
 github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
@@ -959,12 +957,8 @@ k8s.io/kube-openapi v0.0.0-20250304201544-e5f78fe3ede9 h1:t0huyHnz6HsokckRxAF1bY
 k8s.io/kube-openapi v0.0.0-20250304201544-e5f78fe3ede9/go.mod h1:5jIi+8yX4RIb8wk3XwBo5Pq2ccx4FP10ohkbSKCZoK8=
 k8s.io/utils v0.0.0-20241210054802-24370beab758 h1:sdbE21q2nlQtFh65saZY+rRM6x6aJJI8IUa1AmH/qa0=
 k8s.io/utils v0.0.0-20241210054802-24370beab758/go.mod h1:OLgZIPagt7ERELqWJFomSt595RzquPNLL48iOWgYOg0=
-knative.dev/caching v0.0.0-20250117155405-a76aa7cd2bb6 h1:a6oO186Ibm9iBmy9GzJPlJCVJnPPCppwo8NEm12Nnus=
-knative.dev/caching v0.0.0-20250117155405-a76aa7cd2bb6/go.mod h1:xCMZSPoup5BSZ5GQ/Xa8xTEWNIZLLHx9mhPMeREt/ck=
 knative.dev/networking v0.0.0-20250117155906-67d1c274ba6a h1:FaDPXtv42+AkYh/mE269pttPSZ3fDVAjJiEsYUaM4SM=
 knative.dev/networking v0.0.0-20250117155906-67d1c274ba6a/go.mod h1:AIKYMfZydhwXR/60c/3KXEnqEnH6aNEEqulifdqJVcQ=
-knative.dev/operator v0.42.2 h1:wgAWYHwoSFmV+wPHCt5dZahHTHLy2VCM4G82PEo9iSc=
-knative.dev/operator v0.42.2/go.mod h1:cfSpJMgvwmuZ7USaxC+zgEuizMFc/xweREW5DG6J1DA=
 knative.dev/pkg v0.0.0-20250117084104-c43477f0052b h1:a+gP7Yzu5NmoX2w1p8nfTgmSKF+aHLKGzqYT82ijJTw=
 knative.dev/pkg v0.0.0-20250117084104-c43477f0052b/go.mod h1:bedSpkdLybR6JhL1J7XDLpd+JMKM/x8M5Apr80i5TeE=
 knative.dev/serving v0.44.0 h1:c6TXhoSAI6eXt0/1ET3C69jMWYA4ES9FskSan/fBaac=

pkg/constants/constants.go

@@ -283,11 +283,8 @@ type InferenceServiceProtocol string
 
 // Knative constants
 const (
-	AutoscalerKey               = "autoscaler"
-	AutoscalerInitialScaleKey   = "initial-scale"
 	AutoscalerConfigmapName     = "config-autoscaler"
 	AutoscalerAllowZeroScaleKey = "allow-zero-initial-scale"
-	DefaultKnServingName        = "knative-serving"
 	DefaultKnServingNamespace   = "knative-serving"
 	KnativeLocalGateway         = "knative-serving/knative-local-gateway"
 	KnativeIngressGateway       = "knative-serving/knative-ingress-gateway"
@@ -407,7 +404,6 @@ var (
 	// revisions, which prevents the reconciliation loop to be triggered if the annotations is
 	// configured here are used.
 	ServiceAnnotationDisallowedList = []string{
-		autoscaling.InitialScaleAnnotationKey,
 		autoscaling.MinScaleAnnotationKey,
 		autoscaling.MaxScaleAnnotationKey,
 		StorageInitializerSourceUriInternalAnnotationKey,
@@ -533,7 +529,6 @@ const (
 const (
 	IstioVirtualServiceKind = "VirtualService"
 	KnativeServiceKind      = "Service"
-	KnativeServingKind      = "KnativeServing"
 	HTTPRouteKind           = "HTTPRoute"
 	GatewayKind             = "Gateway"
 	ServiceKind             = "Service"

pkg/controller/v1alpha1/inferencegraph/controller.go

@@ -23,7 +23,6 @@ limitations under the License.
 // +kubebuilder:rbac:groups=serving.knative.dev,resources=services/status,verbs=get;update;patch
 // +kubebuilder:rbac:groups=route.openshift.io,resources=routes,verbs=create;get;update;patch;watch;delete
 // +kubebuilder:rbac:groups=route.openshift.io,resources=routes/status,verbs=get
-// +kubebuilder:rbac:groups=operator.knative.dev,resources=knativeservings,verbs=get;list;watch
 package inferencegraph
 
 import (
@@ -48,7 +47,6 @@ import (
 	"k8s.io/client-go/kubernetes"
 	"k8s.io/client-go/rest"
 	"k8s.io/client-go/tools/record"
-	operatorv1beta1 "knative.dev/operator/pkg/apis/operator/v1beta1"
 	"knative.dev/pkg/apis"
 	knservingv1 "knative.dev/serving/pkg/apis/serving/v1"
 	ctrl "sigs.k8s.io/controller-runtime"
@@ -285,19 +283,7 @@ func (r *InferenceGraphReconciler) Reconcile(ctx context.Context, req ctrl.Reque
 		if !ksvcAvailable {
 			r.Recorder.Event(graph, corev1.EventTypeWarning, "ServerlessModeRejected",
 				"It is not possible to use Serverless deployment mode when Knative Services are not available")
-			return reconcile.Result{Requeue: false}, reconcile.TerminalError(fmt.Errorf("the resolved deployment mode of InferenceGraph '%s' is Serverless, but Knative Services are not available", graph.Name))
-		}
-
-		// Abort if Knative KnativeServings are not available
-		knServingFound, knServingCheckErr := utils.IsCrdAvailable(r.ClientConfig, operatorv1beta1.SchemeGroupVersion.String(), constants.KnativeServingKind)
-		if knServingCheckErr != nil {
-			return reconcile.Result{}, knServingCheckErr
-		}
-
-		if !knServingFound {
-			r.Recorder.Event(graph, corev1.EventTypeWarning, "ServerlessModeRejected",
-				"It is not possible to use Serverless deployment mode when Knative KnativeServings are not available")
-			return reconcile.Result{Requeue: false}, reconcile.TerminalError(fmt.Errorf("the resolved deployment mode of InferenceGraph '%s' is Serverless, but Knative KnativeServings are not available", graph.Name))
+			return reconcile.Result{Requeue: false}, reconcile.TerminalError(fmt.Errorf("the resolved deployment mode of InferenceGraph '%s' is Serverless, but Knative Serving is not available", graph.Name))
 		}
 
 		// Retrieve the allow-zero-initial-scale value from the knative autoscaler configuration.
@@ -306,7 +292,7 @@ func (r *InferenceGraphReconciler) Reconcile(ctx context.Context, req ctrl.Reque
 			return ctrl.Result{}, errors.Wrapf(err, "failed to retrieve the knative autoscaler configuration")
 		}
 
-		knutils.ValidateInitialScaleAnnotation(graph.Annotations, allowZeroInitialScale, r.Log)
+		knutils.ValidateInitialScaleAnnotation(graph.Annotations, allowZeroInitialScale, graph.Spec.MinReplicas, r.Log)
 
 		desired := createKnativeService(graph.ObjectMeta, graph, routerConfig)
 

pkg/controller/v1alpha1/inferencegraph/controller_test.go

@@ -239,6 +239,61 @@ var _ = Describe("Inference Graph controller test", func() {
 				Expect(k8sClient.Create(ctx, ig)).Should(Succeed())
 				defer k8sClient.Delete(ctx, ig)
 
+				actualService := &knservingv1.Service{}
+				Eventually(func() error {
+					return k8sClient.Get(context.TODO(), serviceKey, actualService)
+				}, timeout).
+					Should(Succeed())
+
+				Expect(actualService.Spec.Template.Annotations).NotTo(HaveKey(autoscaling.InitialScaleAnnotationKey))
+			})
+		})
+		When("a Serverless InferenceGraph is created with zero min replicas", func() {
+			It("should use the default initial scale value", func() {
+				// Create configmap
+				configMap := &corev1.ConfigMap{
+					ObjectMeta: metav1.ObjectMeta{
+						Name:      constants.InferenceServiceConfigMapName,
+						Namespace: constants.KServeNamespace,
+					},
+					Data: configs,
+				}
+				Expect(k8sClient.Create(context.TODO(), configMap)).NotTo(HaveOccurred())
+				defer k8sClient.Delete(context.TODO(), configMap)
+
+				// Create InferenceGraph
+				graphName := "initialscale4"
+				expectedRequest := reconcile.Request{NamespacedName: types.NamespacedName{Name: graphName, Namespace: "default"}}
+				serviceKey := expectedRequest.NamespacedName
+				ctx := context.Background()
+				var minScale int32 = 0
+				ig := &v1alpha1.InferenceGraph{
+					ObjectMeta: metav1.ObjectMeta{
+						Name:      serviceKey.Name,
+						Namespace: serviceKey.Namespace,
+						Annotations: map[string]string{
+							"serving.kserve.io/deploymentMode": string(constants.Serverless),
+						},
+					},
+					Spec: v1alpha1.InferenceGraphSpec{
+						MinReplicas: &minScale,
+						Nodes: map[string]v1alpha1.InferenceRouter{
+							v1alpha1.GraphRootNodeName: {
+								RouterType: v1alpha1.Sequence,
+								Steps: []v1alpha1.InferenceStep{
+									{
+										InferenceTarget: v1alpha1.InferenceTarget{
+											ServiceURL: "http://someservice.exmaple.com",
+										},
+									},
+								},
+							},
+						},
+					},
+				}
+				Expect(k8sClient.Create(ctx, ig)).Should(Succeed())
+				defer k8sClient.Delete(ctx, ig)
+
 				actualService := &knservingv1.Service{}
 				Eventually(func() error {
 					return k8sClient.Get(context.TODO(), serviceKey, actualService)
@@ -291,7 +346,7 @@ var _ = Describe("Inference Graph controller test", func() {
 				Expect(k8sClient.Create(context.TODO(), configMap)).NotTo(HaveOccurred())
 				defer k8sClient.Delete(context.TODO(), configMap)
 
-				graphName := "initialscale4"
+				graphName := "initialscale5"
 				expectedRequest := reconcile.Request{NamespacedName: types.NamespacedName{Name: graphName, Namespace: "default"}}
 				serviceKey := expectedRequest.NamespacedName
 				ctx := context.Background()
@@ -346,7 +401,7 @@ var _ = Describe("Inference Graph controller test", func() {
 				Expect(k8sClient.Create(context.TODO(), configMap)).NotTo(HaveOccurred())
 				defer k8sClient.Delete(context.TODO(), configMap)
 
-				graphName := "initialscale5"
+				graphName := "initialscale6"
 				expectedRequest := reconcile.Request{NamespacedName: types.NamespacedName{Name: graphName, Namespace: "default"}}
 				serviceKey := expectedRequest.NamespacedName
 				ctx := context.Background()
@@ -401,7 +456,7 @@ var _ = Describe("Inference Graph controller test", func() {
 				Expect(k8sClient.Create(context.TODO(), configMap)).NotTo(HaveOccurred())
 				defer k8sClient.Delete(context.TODO(), configMap)
 
-				graphName := "initialscale6"
+				graphName := "initialscale7"
 				expectedRequest := reconcile.Request{NamespacedName: types.NamespacedName{Name: graphName, Namespace: "default"}}
 				serviceKey := expectedRequest.NamespacedName
 				ctx := context.Background()
@@ -443,6 +498,60 @@ var _ = Describe("Inference Graph controller test", func() {
 				Expect(actualService.Spec.Template.Annotations).NotTo(HaveKey(autoscaling.InitialScaleAnnotationKey))
 			})
 		})
+		When("a Serverless InferenceGraph is created with zero min replicas", func() {
+			It("should override the default initial scale value with zero", func() {
+				// Create configmap
+				configMap := &corev1.ConfigMap{
+					ObjectMeta: metav1.ObjectMeta{
+						Name:      constants.InferenceServiceConfigMapName,
+						Namespace: constants.KServeNamespace,
+					},
+					Data: configs,
+				}
+				Expect(k8sClient.Create(context.TODO(), configMap)).NotTo(HaveOccurred())
+				defer k8sClient.Delete(context.TODO(), configMap)
+
+				graphName := "initialscale8"
+				expectedRequest := reconcile.Request{NamespacedName: types.NamespacedName{Name: graphName, Namespace: "default"}}
+				serviceKey := expectedRequest.NamespacedName
+				ctx := context.Background()
+				var minScale int32 = 0
+				ig := &v1alpha1.InferenceGraph{
+					ObjectMeta: metav1.ObjectMeta{
+						Name:      serviceKey.Name,
+						Namespace: serviceKey.Namespace,
+						Annotations: map[string]string{
+							"serving.kserve.io/deploymentMode": string(constants.Serverless),
+						},
+					},
+					Spec: v1alpha1.InferenceGraphSpec{
+						MinReplicas: &minScale,
+						Nodes: map[string]v1alpha1.InferenceRouter{
+							v1alpha1.GraphRootNodeName: {
+								RouterType: v1alpha1.Sequence,
+								Steps: []v1alpha1.InferenceStep{
+									{
+										InferenceTarget: v1alpha1.InferenceTarget{
+											ServiceURL: "http://someservice.exmaple.com",
+										},
+									},
+								},
+							},
+						},
+					},
+				}
+				Expect(k8sClient.Create(ctx, ig)).Should(Succeed())
+				defer k8sClient.Delete(ctx, ig)
+
+				actualService := &knservingv1.Service{}
+				Eventually(func() error {
+					return k8sClient.Get(context.TODO(), serviceKey, actualService)
+				}, timeout).
+					Should(Succeed())
+
+				Expect(actualService.Spec.Template.Annotations[autoscaling.InitialScaleAnnotationKey]).To(Equal("0"))
+			})
+		})
 	})
 
 	Context("When creating an inferencegraph with headers in global config", func() {

pkg/controller/v1alpha1/inferencegraph/suite_test.go

@@ -30,8 +30,6 @@ import (
 	"k8s.io/client-go/kubernetes"
 	"k8s.io/client-go/kubernetes/scheme"
 	"k8s.io/client-go/rest"
-	"knative.dev/operator/pkg/apis/operator/base"
-	operatorv1beta1 "knative.dev/operator/pkg/apis/operator/v1beta1"
 	knservingv1 "knative.dev/serving/pkg/apis/serving/v1"
 
 	ctrl "sigs.k8s.io/controller-runtime"
@@ -134,24 +132,6 @@ var _ = BeforeSuite(func() {
 	}
 	Expect(k8sClient.Create(context.Background(), configAutoscaler)).Should(Succeed())
 
-	// Create knativeserving custom resource
-	knativeCr := &operatorv1beta1.KnativeServing{
-		ObjectMeta: metav1.ObjectMeta{
-			Name:      constants.DefaultKnServingName,
-			Namespace: constants.DefaultKnServingNamespace,
-		},
-		Spec: operatorv1beta1.KnativeServingSpec{
-			CommonSpec: base.CommonSpec{
-				Config: base.ConfigMapData{
-					"autoscaler": map[string]string{
-						"allow-zero-initial-scale": "true",
-					},
-				},
-			},
-		},
-	}
-	Expect(k8sClient.Create(context.Background(), knativeCr)).Should(Succeed())
-
 	deployConfig := &v1beta1.DeployConfig{DefaultDeploymentMode: "Serverless"}
 
 	err = (&InferenceGraphReconciler{